On the location of the zeros of the derivative of Dirichlet L-functions

Transcription

1 Annals of the University of Bucharest (mathematical series) 5 (LXIII) (2014), On the location of the zeros of the derivative of Dirichlet L-functions Ilie Barza, Dorin Ghisa and Florin Alan Muscutar Dedicated to Professor Cabiria Andreian Cazacu on the occasion of her 85th birthday Abstract - A connection between the Riemann Hypothesis and the distribution of the zeros of the derivative ζ of the Riemann Zeta function ζ has been revealed 80 years ago, but it remained dormant until recently, when a lot of studies began to be devoted to the distribution of the zeros of the derivatives of ζ. Although the first ideas were based on geometric grounds, the recent studies have more a number-theoretical flavour. We revive in this paper the initial geometrical ideas and study the problem in the larger context of the family of Dirichlet L-functions, to which the Riemann Zeta function belongs. Key words and phrases : Dirichlet character, Dirichlet L-function, non trivial zero, fundamental domain. Mathematics Subject Classification (2010) : 30C35, 11M Introduction The idea of using the pre-image of the real axis (die Methode der reellen Züge) in the study of the location of the zeros of ζ is present in Speiser s paper [13] who was inspired by the thesis of A. Utzinger, Über die reellen Züge der Riemannschen Zetafunction, Zürich However, the lack of modern computational tools did not allow him to fully benefit of the power of this method. In particular, it was impossible at that time to find the fundamental domains of the Riemann Zeta function, which are essential in the visualization of the conformal mapping realized by this function. This has been accomplished in [3]-[5] and [7]-[9]. Moreover, in [6] we succeeded to do the same thing for arbitrary Dirichlet L-functions. Let us present the concepts we need in order to tackle this problem. A Dirichlet character modulo q, with q a positive integer, is a totally multiplicative periodic function χ of period q defined on Z (see [11], [6]). A Dirichlet L-series is a series of the form L(s ; χ) = 21 n=1 χ(n) n s. (1.1)

2 22 Ilie Barza, Dorin Ghisa and Florin Alan Muscutar It is known (see [11]) that every Dirichlet L-series can be extended to an analytic function in the whole plane, except possibly at s = 1, where it can have a simple pole. Such a function is called Dirichlet L-function. The Riemann Zeta function is the particular Dirichlet L-function obtained for q = 1. If d q and χ is a Dirichlet character modulo d, then χ(n) = χ (n) if (n, q) = 1 and χ(n) = 0 otherwise, defines a character modulo q for which ( L(s ; χ) = L(s ; χ ) 1 χ(p) ) p s, (1.2) p q,p>1 where p are prime numbers. We say that χ induces χ. A Dirichlet character χ is called primitive if it is not induced by any other character. The formula (1.2) shows that if χ induces χ, then L(s ; χ) and L(s ; χ ) have the same zeros, except for the zeros of the product, which are all on the imaginary axis. We consider them as trivial zeros of L(s ; χ) and therefore L(s ; χ) and L(s ; χ ) have the same non trivial zeros. We would like to do a similar classification of the zeros of the derivative L (s ; χ). This was straightforward for the derivative of the Riemann Zeta function for which the real zeros were considered naturally as trivial, while the others were considered non trivial. A criterion of triviality for the zeros of L (s ; χ) could come from the way these zeros are generated by the zeros of L(s ; χ). Indeed, we have seen in [6] that for any bounded region D of the plane, we can find r > 0 such that the intersection of D with the pre-image of a circle γ r centred at the origin and of radius r is a set of disjoint closed curves containing each one a unique zero of L(s ; χ) situated in D. When r increases, these curves expand and for some value r 0 of r two of them touch each other at a point v 0. This point is a branch point of the branched Riemann surface generated by L(s ; χ) and therefore a zero of L (s ; χ). By letting r increase past r 0 the two curves fuse into a unique closed curve containing both zeros and for a greater value of r some other zeros of L (s ; χ) are obtained and so on. For the Riemann Zeta function, the trivial zeros of ζ have generated in this way the trivial zeros of ζ and the non trivial zeros of ζ have generated the non trivial zeros of ζ. The situation becomes more complicated for arbitrary Dirichlet L-functions, since trivial zeros on the imaginary axis and non trivial ones can generate together zeros of its derivative. Which ones of these zeros will be called trivial and which ones non trivial? In order to device a proper criterion, we need to remember (see [6] and [8]) that all the zeros of L(s ; χ) and L (s ; χ) are simple zeros. Moreover, there are unbounded strips S k, k 0 bounded each one by a couple of unbounded curves which are mapped bijectively by L(s ; χ) onto the interval (1, + ) of the real axis such that every S k contains j k 1 non trivial zeros located on the critical axis Re s = 1/2 and a number m k 0 of trivial zeros located on the imaginary axis. There is also a strip S 0 bounded by the same type of curves which contains all the trivial real zeros of L(s ; χ). The zeros of L (s ; χ) are obtained by expanding the components

3 On the location of the zeros of the derivative of Dirichlet L-functions 23 of the pre-image by L(s ; χ) of γ r. If both components touching at a v k,j contain non trivial zeros of L(s ; χ), we will say that v k,j is a non trivial zero of L (s ; χ), otherwise we will call it trivial. This definition agrees with the known definition of the (non)triviality of the zeros of the Riemann Zeta function. It is an easy exercise to check that S k contains j k 1 non trivial zeros and m k trivial zeros of L (s ; χ), where j k and m k are the non trivial and respectively trivial zeros of L(s ; χ) from S k. These numbers agree with those known for Dirichlet L-functions defined by primitive characters, in which case m = 0. They are obtained by building a complete binary tree having as leaves the zeros of L(s ; χ) from S k and as internal nodes the zeros of L (s ; χ) from S k. It is known that if such a tree has j k + m k leaves, it will have j k + m k 1 internal nodes. From these j k 1 correspond to the non trivial zeros of L (s ; χ), as it can be seen by building separately the tree generated by the non trivial zeros of L(s ; χ), i.e. by simply ignoring the existence of trivial zeros of L(s ; χ). We call progenitor of a leaf an internal node from which one can descend to that leaf. Figure 1 below illustrates the strips S k as well as the components of the pre-images of circles γ r for different values of r in the case of a Dirichlet L-function defined by a complex Dirichlet character χ modulo 14, which is imprimitive and thus L(s ; χ) has some imaginary trivial zeros. The way the zeros of L(s ; χ) generate the zeros of L (s ; χ) is shown in the adjacent binary tree. We can distinguish trivial and non trivial zeros of L(s ; χ) as well as of L (s ; χ). A lot of studies (see [10]-[14]) have been devoted lately to the distribution of the zeros of the derivatives of ζ and most of them make reference to Speiser s paper without expanding on the method he used. We take in this paper a deeper look into the geometry of the pre-image of the real axis in a more general setting, namely for an arbitrary Dirichlet L-function, in order to draw conclusions related to the distribution of the zeros of its derivative. 2. The pre-image of the real axis by L(s ; χ) The alternative notation L(q, j, s) is used for a Dirichlet L-function defined by the j-th Dirichlet character modulo q when it is important to specify both of these parameters. The location of the zeros of such a function becomes obvious when coloring differently the pre images of the negative and of the positive real half axes, namely the zeros appear at the junction of the two colors. By the Big Picard Theorem, if x 0 R is a non lacunary value of L(q, j, s), then there is a countable number of points s n C such that L(q, j, s n ) = x 0. The continuation along R from every s n (see [2]) produces a countable number of unbounded curves, called components of the pre-image of R by L(q, j, s). Sometimes, in the continuation process, branch points of the respective branched Riemann surface can be met. These are the points where L (q, j, s) = 0. It is known (see [1], page 133 ) that in the neighborhood of such a point s 0 we have:

4 24 Ilie Barza, Dorin Ghisa and Florin Alan Muscutar Figure 1. Trivial and non trivial zeros of L(s ; χ) and L (s ; χ). L(q, j, s) = L(q, j, s 0 ) + (s s 0 ) 2 g(s), (2.1) where g(s) is an analytic function and g(s 0 ) 0. Indeed, it is known (see [6]) that the zeros of L (q, j, s) are all simple, hence two components of the pre-image of R by L(q, j, s) touch each other at s 0 and are orthogonal.

5 On the location of the zeros of the derivative of Dirichlet L-functions 25 It has been shown in [6] that for any Dirichlet L-function there are four kinds of components of the pre-image of R. Namely, there are countable many components Γ k which are mapped bijectively by the function onto the interval (1, + ). They do not intersect each other and form infinite strips S k. Every strip S k contains a unique component Γ k,0 of the pre-image of R which is mapped bijectively by the function onto the interval (, 1) and a finite number j k 1 of components Γ k,j, j 0 which are mapped bijectively by the function onto the whole real axis. The curves Γ k,j do not intersect each other and they do not intersect any curve Γ k. Every curve Γ k,j contains a unique zero s k,j of the L-function, which is non trivial if the function is generated by a primitive character and can be a trivial zero on the imaginary axis if the function is generated by an imprimitive one. Finally, there are those components containing trivial real zeros, which belong all to the strip S 0 and which are mapped by the function onto the whole real axis. Adjacent components touch each other orthogonally on the real axis if the character is real. If the character is not real, then they do not touch each other and every strip between two adjacent components contains a unique branch point of the function. We can reach these branch points, as well as those belonging to the strips S k by arcs starting at points u k,j for which L(u k,j ; χ) = 1 or by unbounded curves on which lim σ >+ L(σ + it ; χ) = 1 and such that the images of these arcs and unbounded curves are the segments connecting z = 1 with the image by L(s ; χ) of the respective branch points. The following is in line with Speiser s results. Theorem 2.1. All the non trivial zeros of L (s ; χ) have the real part greater than 1/2.The trivial zeros of L (s ; χ) with imaginary progenitors have the real part greather than 0. Proof. Let v k,j be such a zero of L (s ; χ) and suppose that it is a progenitor of the non trivial zero s k,j. Due to the color alternating rule (see [5], [6]), v k,j is situated outside the parabola-like curve Γ k,j. It is obviously enough to study the case when j > 0, i.e. when Γ k,j is above Γ k,0. Then L(v k,j ; χ) belongs to the upper half plane in Figure 2, hence 0 < α < π, where α = arg L(v k,j ; χ). In this paper arg z [0, 2π) is the angle made by the ray from the origin through z with the positive real half axis. The pre-image of the ray determined by L(v k,j ; χ) contains two curves which are orthogonal at v k,j. Indeed, it is known (see [6]) that the zeros of L (s ; χ) are simple, hence v k,j is a ramification point of order two. The angles at v k,j are doubled by L(s ; χ), hence the four arcs meeting at v k,j make the angles α/2, π/2 + α/2, π + α/2 and 3π/2 + α/2 with a horizontal line. The angle of π/2 + α/2 made by the second arc (which ends at s k,j ) is less than the angle β made by the tangent to it at any other point and the horizontal line. If Re v k,j Re s k,j, then there must be a point on this arc for which β = π/2, therefore π/2 + α/2 < π/2, i.e. α < 0, which is absurd and this contradiction proves the first part of

6 26 Ilie Barza, Dorin Ghisa and Florin Alan Muscutar Figure 2. Re v k,j > Re s k,j. the theorem. The second part is obtained taking Re s k,j = 0. Figure 2 below illustrates such a situation for ζ(s) in the strip S 7 and the non trivial zero s 7,1 = 1/ i of ζ as well as the corresponding zero v 7,1 = i of ζ. 3. The second derivative of L(s ; χ) Let us take now the pre-image by L (s ; χ) of the real axis. Some of the components are this time curves Υ k which bound strips Σ k and are mapped bijectively by L (s ; χ) onto the interval (, 0). Every strip Σ k contains a unique component Υ k,0 of the pre-image of the real axis, which is mapped bijectively by L (s ; χ) onto the interval (0, + ) and a number of components which are mapped bijectively by L (s ; χ) onto R. Let us take also the preimage by L (s ; χ) of a circle γ r centered at the origin and of radius r. Since lim σ >+ L (σ + it ; χ) = 0, for every ɛ > 0, there is σ t such that σ > σ t implies that L (σ + it ; χ) < ɛ. Therefore, for r < ɛ some components of the pre-image of γ r must intersect the unbounded sets {s = σ + it σ > σ t } for every t R. In fact, we can show that there is just one such component. Indeed, by Big Picard Theorem, the pre-image of the point z = r is a countable set of points accumulating to. Each one of these points belongs to a unique component of the pre-image of the negative real half axis. In other words every curve Υ k contains one of these points and so does every curve Υ k,j. Starting from such a point on Υ k and performing continuation along γ r counterclockwise we reach the point on Υ k+1 and if the continuation is clockwise, we reach the point on Υ k 1. It is obvious that in this way we

7 On the location of the zeros of the derivative of Dirichlet L-functions 27 visit all the points belonging to the pre-image of z = r which are situated on curves Υ k and the unbounded curve we obtain in this way is the unique (due to the monodromy theorem) unbounded component of the pre-image of γ r. For an arbitrary bounded region of the plane, we can take r small enough such that the bounded components of the pre-image by L (s ; χ) of γ r included in that region are disjoint closed curves containing each one a zero of L (s ; χ). Indeed, when r > 0 the respective components contract each one to a point, which is a zero of L (s ; χ). Theorem 3.1. The number of zeros of L (s ; χ) and of L (s ; χ) from a given strip Σ k is the same. Proof. Suppose that Σ k contains m zeros of L (s ; χ). Considering them as the leaves of a binary tree whose internal nodes are obtained as the touching points of the bounded components of the pre-image of γ r, we obtain a complete binary tree. It is known that the number of internal nods of this tree must be m 1. These are not all the zeros of L (s ; χ). One more zero is obtained as the touching point of one of these components with the unbounded component of the pre-image of γ r. Every zero of L (s ; χ) in Σ k can be obtained in this way and therefore the number of zeros of L (s ; χ) in Σ k is exactly m. Some of the zeros of L (s ; χ) from Σ k can be trivial zeros, as defined in Section 1. We apply that definition recursively for the zeros of any derivative of L(s ; χ). Therefore, the non trivial zeros of L (s ; χ) are obtained as internal nodes of the complete binary tree built on the non trivial zeros of L (s ; χ) as leaves. The other zeros of L (s ; χ) from Σ k are trivial, even if they are progenitors of some non trivial zeros of L (s ; χ). Theorem 3.2. To every zero v k,j of L (s ; χ) from Σ k corresponds a zero w k,j of L (s ; χ) such that Re w k,j > Re v k,j and Im w k,j Im v k,j. Proof. For a zero w k,j of L (s ; χ), the pre-image by L (s ; χ) of the ray determined by L (w k,j ) has two components orthogonal in w k,j. If arg L (w k,j ; χ) = α, then as in Theorem 2.1 we can deal just with the case 0 < α < π. The four arcs starting at w k,j form with a horizontal line respectively the angles α/2, π/2 + α/2, π + α/2 and 3π/2 + α/2. The second arc ends in v k,j and the tangent to it makes with a horizontal line an angle β > π/2 + α/2. For Re w k,j Re v k,j, there would be a point on this arc for which β = π/2, hence π/2 > π/2 + α/2, i.e. α < 0, which is a contradiction, proving the first part of the theorem. For the second, we need to explain first the meaning of the sign used there. We have seen that lim σ >+ L (σ + it ; χ) = 0 implies the existence of an unbounded component of the pre-image by L (s ; χ) of every circle γ r centered at the origin and of radius r. We can take r small enough such that this component and

8 28 Ilie Barza, Dorin Ghisa and Florin Alan Muscutar the bounded components from a given S k are all disjoint. Letting r increase, the bounded components expand, remaining in S k. Indeed, due to the color alternating rule, they cannot cut Γ k or Γ k+1. The unbounded component moves indefinitely to the left with the increasing r, since for σ + it Υ k we have that lim σ > L (σ + it ; χ) =. There must be a value of r for which the bounded component containing v k,j meets the unbounded component of the pre-image of γ r at a point w k,j. Since the last does not differ too much in Σ k of a vertical line, this explains the fact that the imaginary parts of v k,j and w k,j do not differ too much. If we continue to increase r the respective curves will fuse into a unique unbounded curve leaving both v k,j and w k,j at the right. Let us notice that there is a hint of the correspondence between s k,j, v k,j and w k,j for the Riemann Zeta function in [14], Figure 1, and Table II, except that the idea of a mysterious bouncing effect (see [14], p. 680) is hiding its true nature. Moreover, the existence of the non trivial zero i of ζ, as Spira notices, contradicts either the generality of this correspondence, or Speiser s results. Indeed, either this zero does not correspond to a non trivial zero of ζ, contradicting the generality of the respective correspondence, or there is a non trivial zero of ζ at the left of the critical line and therefore Speiser s result is not true, admitting the Riemann Hypothesis. However, this correspondence concerns the non trivial zeros of ζ, ζ and ζ and for a meaningful discussion on this topic, we should first clarify the nature of the zero i. We anticipate by saying that it is a trivial zero of ζ and the whole problem can be now looked upon from a different perspective. It is also worth to compare at this moment the effectiveness of the number-theoretical method used by Spira and that of the geometric function theory consistently used by us in [3]-[9]. The data regarding the non trivial zeros of ζ and ζ from [14], Figure 1 are obtained in [4] and [8] by taking the pre-image by ζ of the real axis, as well as of some circles γ r centered at the origin. The first pre-image partitions the complex plane into strips S k in which the second pre-image operates locating the zeros of ζ. In every strip S k the number of these zeros is one unit less than those of ζ. The reason is the way they appear as the internal nodes of a complete binary tree whose leaves are the zeros of ζ contained by the respective strip. This difference explains the gaps noticed by Spira and labelled bouncing. Moreover, starting now with the zeros of ζ from a given strip S k, the color matching rule established in [7] allows us to substitute Σ k with S k and obtain in the same way the zeros of ζ situated in S k, except that this time there is an unbounded component of the pre-image of γ k, creating a zero of ζ, such that the number of zeros of ζ in S k is the same as that of ζ. Moreover, it appears that these zeros are at the intersection of the unbounded component of the pre-image of γ r with the bounded ones, which explains the fact noticed by Spira that Im w k,j is approximately the same with Im v k,j. Having this in

9 On the location of the zeros of the derivative of Dirichlet L-functions 29 Figure 3. The correspondence of the zeros of ζ, ζ and ζ. mind, we can complete, as shown in Figure 3a below, Spira s Figure 1 ([14], p. 681), by indicating the strips S k, the corresponding complete binary trees of the zeros of ζ and ζ, as well as the correspondence v k,j > w k,j. The Figure 3 above shows the non trivial zeros of ζ, ζ and ζ as calculated by Spira, but also as deduced by our geometric method. We state as a theorem the surprising result regarding the zero i. Theorem 3.3. The function ζ has two non real trivial zeros symmetric with respect to the real axis. Proof. Indeed, a trivial zero of ζ is obtained when two components of a pre-image by ζ (s) of a ray cross each other and at least one of them ends in

10 30 Ilie Barza, Dorin Ghisa and Florin Alan Muscutar Figure 4. Non real trivial zeros of ζ. a trivial zero of ζ. This is exactly the situation with the zero i of ζ. Indeed, the pre-image of the ray passing through ζ ( i) contains two curves passing through the pole s = 1 and crossing each other once more at i (see Figure 4b). One of the components Γ of this pre-image passes through the first trivial zero of ζ. Let us notice that lim σ >+ ζ (σ +it) = 0 when σ +it Γ. The other component Γ stretches for σ from - to + and lim σ >+ ζ (σ+it) = 0 and lim σ > ζ (σ+it) = when σ + it Γ. We notice also that the same zero i is obtained as a touching point of two components of the pre-image by ζ (s) of the circle γ r of radius r = ζ ( i) and centered at the origin, one bounded and the other one unbounded (see Figure 4a). We varied slightly the value of r to show the way these components approach each other, touch each other and then fuse to a unique unbounded component. Figure 4c reveals the existence of the trivial zero i of ζ. It also illustrates the fact that s = 1 is a double pole for ζ (blue and yellow) and a triple pole for ζ (red and black). In our knowledge, the question whether the zeros of L (n) (s ; χ) have the real part as big as we want for n big enough, was not answered yet. Spira s affirmation that it seems probable that the zeros of ζ (k) (s) will have real parts tending to with k has no solid support in his paper and we did not find a similar affirmation elsewhere. All we can say based on Theorem 3.2 is that if σ n + it n is a zero of L (n) (s ; χ), then there is a zero σ n+1 +it n+1 of L (n+1) (s ; χ) such that σ n+1 > σ n. References [1] L.V. Ahlfors, Complex Analysis, International Series in Pure and Applied Mathematics, 1979.

11 On the location of the zeros of the derivative of Dirichlet L-functions 31 [2] L.V. Ahlfors and L. Sario, Riemann Surfaces, Princeton University Press, [3] C. Andreian Cazacu and D. Ghisa, Global mapping properties of analytic functions, Proceedings of 7-th ISAAC Congress, London, U.K., pp. 3-12, 2009, [4] C. Andreian Cazacu and D. Ghisa, Fundamental Domains of Gamma and Zeta Functions, Int. J. Math. Math. Sci., ID , 21 pages, 2011, [5] C. Ballantine and D. Ghisa, Colour visualization of Blaschke product mappings, Complex Var. Elliptic Equ., 55 (2010), , [6] D. Ghisa, On the Generalized Riemann Hypothesis, to appear. [7] D. Ghisa, Fundamental Domains and the Riemann Hypothesis, Lambert Academic Publishing, [8] D. Ghisa and F.A. Muscutar, The Riemann Zeta function has only simple zeros, Int. J. Math. Sci. Eng. Appl., 6 (2012), [9] D. Ghisa and G.S. Salagean, Distortion and related inequalities for the Riemann Zeta function, Int. J. Math. Sci. Eng. Appl., 6 (2012), [10] N. Levinson and H.L. Montgomery, Zeros of the derivative of the Riemann Zeta function, Acta Math., 133 (1974), 49-65, [11] H.L. Montgomery and R.C. Vaughan, Multiplicative Number Theory, Cambridge Studies in Advanced Mathematics, 97, 2007, [12] K. Soundararajan, The horizontal distribution of the zeros of ζ (s), Duke Math. J., 91 (1998), 33-59, [13] A. Speiser, Geometrisches zur Riemannsche Zetafunktion, Math. Ann., 110 (1935), , [14] R. Spira, Zero-free regions of ζ (k) (s), J. Lond. Math. Soc., 40 (1965), , Ilie Barza Karlstad University, Dept. of Engineering Sciences, Physics and Mathematics S-651-Karlstad, Sweden Sorina.Barza@kau.se Dorin Ghisa York University, Glendon College, Department of Mathematics 2275 Bayview Avenue, Toronto, Canada, M4N 3M6 dghisa@yorku.ca Florin Alan Muscutar Lorain CC College, Science and Mathematics 1005 N Abbe Road, Elyria, OH fmuscuta@lorainccc.edu